Skip to content

Fix attn_implementation name in OnlineDPO for transformers v5 #4322

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 22, 2025
Merged

Fix attn_implementation name in OnlineDPO for transformers v5 #4322

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions trl/trainer/online_dpo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data
import transformers
from accelerate import logging
from accelerate.utils import broadcast_object_list, gather_object, is_peft_model
from datasets import Dataset
Expand Down Expand Up @@ -1049,10 +1050,11 @@ def _generate(self, model, prompts, images=None):
if self.use_transformers_paged:
previous_attn = self.model_wrapped.config._attn_implementation

if is_flash_attn_2_available():
self.model_wrapped.config._attn_implementation = "paged_attention"
if version.parse(transformers.__version__).release >= version.parse("5.0.0").release:
new_attn = "paged|flash_attention_2" if is_flash_attn_2_available() else "paged|sdpa"
else:
self.model_wrapped.config._attn_implementation = "sdpa_paged"
new_attn = "paged_attention" if is_flash_attn_2_available() else "sdpa_paged"
self.model_wrapped.config._attn_implementation = new_attn
with (
profiling_context(self, "transformers.generate_batch"),
unwrap_model_for_generation(
Expand Down
Loading